Research On Water Quality Prediction In Shanghai Based On CEEMDAN-LSTM Model

Excessive amounts of ammonia nitrogen, total phosphorus, and total nitrogen in surface water can cause serious water pollution problems. Accurate prediction of changes in these three indicators is of great significance for water resource protection. However, the accuracy of existing prediction models is not high, and water quality detection data exhibits characteristics such as nonlinearity, large fluctuations, and long periods. Under the background of polluted water resources, a CEEMDAN-LSTM hybrid model is proposed in this paper, using the water quality detection data of the Sanjia Port section in Pudong New District, Shanghai as the sample data. First, CEEMDAN is used to smooth the surface water data. Afterwards, the final prediction is performed by the LSTM model. Water quality test indicators include ammonia nitrogen, total phosphorus, and total nitrogen. CNN, SVR, GRU and LSTM are adopted as benchmark models for comparison in this research. The research results demonstrate that the CEEMDAN-LSTM model's prediction root mean square errors for NH3-H, TP, and TN are 0.065, 0.038, and 0.127 respectively, with the smallest deviation degree. The prediction accuracy is high, which illustrates the rationality of the model. This model provides effective technical support for water quality prediction in the future.